采用SIFT特征的空基动态视频稳定技术  

Stabilization algorithm based on SIFT feature for dynamic airborne video

作　　者：林锦梅[1] 周付根[1] 金挺[1] 

机构地区：[1]北京航空航天大学宇航学院图像处理中心,北京100191

出　　处：《红外与激光工程》2011年第12期2552-2557,共6页Infrared and Laser Engineering

基　　金：国家863计划(2006AA11Z232)

摘　　要：当采用空基平台对道路进行交通检测时,平台沿道路飞行,空基平台姿态不易控制且控制精度较低,此外,由于风速影响和平台自身振动等因素,检测获取的视频图像存在不必要的随机摇摆和抖动,为了去除抖动,改善观测效果,需进行动态观测模式下的视频稳定处理以实现稳定观测。采用改进的SIFT算法进行特征提取,提高了SIFT特征提取的效率,并根据动态视频相邻帧匹配的实际应用,采用邻域搜索方法进行特征匹配,提高了匹配的精度。得到精确匹配的特征点对进行运动参数估计,并采用Kalman滤波对运动参数平滑后进行视频图像的校正补偿,得到稳定的视频输出。该算法精度较高,稳定效果较好,能有效地实现空基平台动态视频稳定处理,便于交通监控,为后续的目标检测与跟踪提供了便利。When the airborne platform is flying along the road for traffic detection, it is not easy to control the attitude of platform and the control accuracy is always low. In addition, the captured video images exist unexpected random swaying and jitter caused by the wind speed effects and the shake of the platform. In order to remove the jitter, video stabilization is needed for improving observing results. An improved algorithm based on SIPT feature was applied for feature extraction and the speed of feature extraction was raised. Neighborhood search which improved the accuracy was used for feature match of inter-frame sequence. After accurate pairs of points were gained for motion estimation, the motion parameters were smoothed by Kalman filter for following motion correction and compensation to get stabilized video sequence. The method is precise enough to obtain a stabilized video for the aerial monitoring, traffic detection and object tracking.

关 键 词：视频稳定 SIFT特征 运动估计 KALMAN滤波 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]